Longitudinal vs Transverse Cellular Orientation in Leiomyosarcoma
In this mini-project, we will assess major-minor axis ratio and alignment of nuclei in H&E histopathology images of Soft Tissue Leiomyosarcoma
# first let's see an image with cells in longitudinal orientation
longitudinal_image_file = 'longitudinal.png'
long_im_input = skimage.io.imread(longitudinal_image_file)[:,:,:3]
plt.imshow(long_im_input)
_ = plt.title('Longitudinal Cellular Orientation', fontsize=16)
transverse_image_file = 'transverse.png'
trans_im_input = skimage.io.imread(transverse_image_file)[:,:,:3]
plt.imshow(trans_im_input)
_ = plt.title('Transverse Cellular Orientation', fontsize=16)
ref_image_file = ('6070-7712.png') # L1.png
im_reference = skimage.io.imread(ref_image_file)[:,:,:3]
# get mean and stddev of reference image in lab space
mean_ref, std_ref = htk.preprocessing.color_conversion.lab_mean_std(im_reference)
long_im_nmzd = htk.preprocessing.color_normalization.reinhard(long_im_input, mean_ref, std_ref)
trans_im_nmzd = htk.preprocessing.color_normalization.reinhard(trans_im_input, mean_ref, std_ref)
plt.figure(figsize=(20, 10))
plt.subplot(1, 2, 1)
plt.imshow(im_reference)
_ = plt.title('Reference Image', fontsize=titlesize)
plt.subplot(1, 2, 2)
plt.imshow(long_im_nmzd)
_ = plt.title('Normalized Longitudinal Input Image', fontsize=titlesize)
plt.figure(figsize=(20, 10))
plt.subplot(1, 2, 1)
plt.imshow(im_reference)
_ = plt.title('Reference Image', fontsize=titlesize)
plt.subplot(1, 2, 2)
plt.imshow(trans_im_nmzd)
_ = plt.title('Normalized Transverse Input Image', fontsize=titlesize)
# create stain to color map
stainColorMap = {
'hematoxylin': [0.65, 0.70, 0.29],
'eosin': [0.07, 0.99, 0.11],
'dab': [0.27, 0.57, 0.78],
'null': [0.0, 0.0, 0.0]
}
# specify stains of input image
stain_1 = 'hematoxylin' # nuclei stain
stain_2 = 'eosin' # cytoplasm stain
stain_3 = 'null' # set to null if input contains only two stains
# create stain matrix
W = np.array([stainColorMap[stain_1],
stainColorMap[stain_2],
stainColorMap[stain_3]]).T
long_im_stains = htk.preprocessing.color_deconvolution.color_deconvolution(long_im_nmzd, W).Stains
# Display results
plt.figure(figsize=(20, 10))
plt.subplot(1, 2, 1)
plt.imshow(long_im_stains[:, :, 0])
plt.title(stain_1, fontsize=titlesize)
plt.subplot(1, 2, 2)
plt.imshow(long_im_stains[:, :, 1])
_ = plt.title(stain_2, fontsize=titlesize)
trans_im_stains = htk.preprocessing.color_deconvolution.color_deconvolution(trans_im_nmzd, W).Stains
# Display results
plt.figure(figsize=(20, 10))
plt.subplot(1, 2, 1)
plt.imshow(trans_im_stains[:, :, 0])
plt.title(stain_1, fontsize=titlesize)
plt.subplot(1, 2, 2)
plt.imshow(trans_im_stains[:, :, 1])
_ = plt.title(stain_2, fontsize=titlesize)
# get nuclei/hematoxylin channel
long_im_nuclei_stain = long_im_stains[:, :, 0]
trans_im_nuclei_stain = trans_im_stains[:, :, 0]
# segment foreground
foreground_threshold = 60
long_im_fgnd_mask = sp.ndimage.morphology.binary_fill_holes(
long_im_nuclei_stain < foreground_threshold)
trans_im_fgnd_mask = sp.ndimage.morphology.binary_fill_holes(
trans_im_nuclei_stain < foreground_threshold)
# run adaptive multi-scale LoG filter
min_radius = 10
max_radius = 15
long_im_log_max, long_im_sigma_max = htk.filters.shape.cdog(
long_im_nuclei_stain, long_im_fgnd_mask,
sigma_min=min_radius * np.sqrt(2),
sigma_max=max_radius * np.sqrt(2)
)
trans_im_log_max, trans_im_sigma_max = htk.filters.shape.cdog(
trans_im_nuclei_stain, trans_im_fgnd_mask,
sigma_min=min_radius * np.sqrt(2),
sigma_max=max_radius * np.sqrt(2)
)
# detect and segment nuclei using local maximum clustering
local_max_search_radius = 10
long_im_nuclei_seg_mask, seeds, maxima = htk.segmentation.nuclear.max_clustering(
long_im_log_max, long_im_fgnd_mask, local_max_search_radius)
trans_im_nuclei_seg_mask, seeds, maxima = htk.segmentation.nuclear.max_clustering(
trans_im_log_max, trans_im_fgnd_mask, local_max_search_radius)
# filter out small objects
min_nucleus_area = 80
long_im_nuclei_seg_mask = htk.segmentation.label.area_open(
long_im_nuclei_seg_mask, min_nucleus_area).astype(np.int)
trans_im_nuclei_seg_mask = htk.segmentation.label.area_open(
trans_im_nuclei_seg_mask, min_nucleus_area).astype(np.int)
# compute nuclei properties
LongObjProps = skimage.measure.regionprops(long_im_nuclei_seg_mask)
TransObjProps = skimage.measure.regionprops(trans_im_nuclei_seg_mask)
print('Number of nuclei in longitudinal image = ', len(LongObjProps))
print('Number of nuclei in transverse image = ', len(TransObjProps))
plt.figure(figsize=(20, 10))
plt.subplot(1, 2, 1)
plt.imshow(skimage.color.label2rgb(long_im_nuclei_seg_mask, long_im_input, bg_label=0), origin='lower')
plt.title('Nuclei segmentation mask overlay', fontsize=titlesize)
plt.subplot(1, 2, 2)
plt.imshow( long_im_input )
plt.xlim([0, long_im_input.shape[1]])
plt.ylim([0, long_im_input.shape[0]])
plt.title('Nuclei bounding boxes', fontsize=titlesize)
for i in range(len(LongObjProps)):
c = [LongObjProps[i].centroid[1], LongObjProps[i].centroid[0], 0]
width = LongObjProps[i].bbox[3] - LongObjProps[i].bbox[1] + 1
height = LongObjProps[i].bbox[2] - LongObjProps[i].bbox[0] + 1
cur_bbox = {
"type": "rectangle",
"center": c,
"width": width,
"height": height,
}
plt.plot(c[0], c[1], 'g+')
mrect = mpatches.Rectangle([c[0] - 0.5 * width, c[1] - 0.5 * height] ,
width, height, fill=False, ec='g', linewidth=2)
plt.gca().add_patch(mrect)
plt.figure(figsize=(20, 10))
plt.subplot(1, 2, 1)
plt.imshow(skimage.color.label2rgb(trans_im_nuclei_seg_mask, trans_im_input, bg_label=0), origin='lower')
plt.title('Nuclei segmentation mask overlay', fontsize=titlesize)
plt.subplot(1, 2, 2)
plt.imshow( trans_im_input )
plt.xlim([0, trans_im_input.shape[1]])
plt.ylim([0, trans_im_input.shape[0]])
plt.title('Nuclei bounding boxes', fontsize=titlesize)
for i in range(len(TransObjProps)):
c = [TransObjProps[i].centroid[1], TransObjProps[i].centroid[0], 0]
width = TransObjProps[i].bbox[3] - TransObjProps[i].bbox[1] + 1
height = TransObjProps[i].bbox[2] - TransObjProps[i].bbox[0] + 1
cur_bbox = {
"type": "rectangle",
"center": c,
"width": width,
"height": height,
}
plt.plot(c[0], c[1], 'g+')
mrect = mpatches.Rectangle([c[0] - 0.5 * width, c[1] - 0.5 * height] ,
width, height, fill=False, ec='g', linewidth=2)
plt.gca().add_patch(mrect)
LongObjPropsTable = skimage.measure.regionprops_table(long_im_nuclei_seg_mask,
properties=('label',
'orientation',
'major_axis_length',
'minor_axis_length',)
)
TransObjPropsTable = skimage.measure.regionprops_table(trans_im_nuclei_seg_mask,
properties=('label',
'orientation',
'major_axis_length',
'minor_axis_length')
)
long_orientation = LongObjPropsTable['orientation']
trans_orientation = TransObjPropsTable['orientation']
_ = plt.hist(x = (long_orientation, trans_orientation),
density=True,
)
plt.xlabel('Angle between the 0th axis and the major axis of the ellipse that has the same second moments as the region')
plt.title("Orientation of major axis of nuclei")
plt.legend(['Cells in longitudinal orientation',
'Cells in transverse orientation'])
ax = plt.gca()
ax.xaxis.set_major_locator(plt.MultipleLocator(np.pi / 2))
ax.xaxis.set_major_formatter(plt.FuncFormatter(multiple_formatter()))
plt.show()
long_minor = LongObjPropsTable['minor_axis_length']
trans_minor = TransObjPropsTable['minor_axis_length']
long_major = LongObjPropsTable['major_axis_length']
trans_major = TransObjPropsTable['major_axis_length']
long_major_minor_ratio = long_major / long_minor
trans_major_minor_ratio = trans_major / trans_minor
_ = plt.hist(x = (long_major_minor_ratio, trans_major_minor_ratio),
density=True,
)
plt.xlabel("Major-minor axis ratio of nuclei")
plt.title("Major-minor axis ratio of nuclei")
plt.legend(['Cells in longitudinal orientation',
'Cells in transverse orientation'])
plt.show()
